New Rockets Missed Their Promised Dates by 10 Months. Probes Mostly Didn't.

On 2018-06-27, NASA set a launch date of 2021-03-30 for the James Webb Space Telescope [3]. Webb launched on 2021-12-25 [2]. That is 270 days late. I love Webb, and I still count the 270 days.

My working claim was this: for large crewed and uncrewed missions, the median launch comes more than 90 days after the date promised about two years before. I also expected agency programs to slip at least as much as commercial ones. I checked both claims against a sourced ledger of 12 missions. The first claim survives, but only weakly. The second claim fails in this sample. The largest effect is a split that I did not plan to test. In the median case, science spacecraft launched on time or early. New rockets and new crew ships launched about 10 months late.

Question

Which date was promised, and how far did the real launch move from it? I want a number in days, with its spread. I do not want a general feeling of "delays".

Data and where it came from

For each mission I need two dates. The first is an official dated statement of a target or commitment. I took it from the agency, the company or a report that quotes their briefing, and it is usually 15 to 42 months before launch. The second is the real launch date in UTC. GCAT, Jonathan McDowell's catalog of every orbital launch, is my independent launch record [2]. I also read the GAO's July 2026 assessment of NASA's major projects for commitment dates and Artemis II [1].

Mission	Owner of the program	Promise and date of statement	Promised date used	Real launch (UTC)	Lead (months)	Slip (days)
Roman Space Telescope	NASA	"by May 2027" commitment [8]	2027-05-31	2026-08-30 [2]	n/a	-274
IMAP	NASA	committed launch readiness December 2025 [1]	2025-12-31	September 2025 [1]	n/a	-92 (month level)
Europa Clipper	NASA	October 2024, stated 2021-07-23 [5]	2024-10-31	2024-10-14 [2]	39	-17
Lucy	NASA	October 2021, stated 2019-01-31 [6]	2021-10-31	2021-10-16 [2]	33	-15
DART	NASA	June 2021, stated 2019-04-12 [7]	2021-06-30	2021-11-24 [2]	31	147
Artemis II	NASA	September 2025 [1]	2025-09-30	2026-04-01 [2]	n/a	183
Ariane 6 first flight	ESA	fourth quarter 2023, stated 2022-10-19 [12]	2023-12-31	2024-07-09 [2]	21	191
James Webb Space Telescope	NASA	2021-03-30, stated 2018-06-27 [3]	2021-03-30	2021-12-25 [2]	42	270
Vulcan first flight	ULA	first quarter 2023, stated 2022-10-12 [11]	2023-03-31	2024-01-08 [2]	15	283
Artemis I	NASA	"by November 2021", stated 2020-08-26 [4]	2021-11-30	2022-11-16 [2]	27	351
Crew Dragon Demo-2	SpaceX (NASA contract)	April 2019, stated 2018-08-02 [9]	2019-04-30	2020-05-30 [2]	22	396
Starliner Crew Flight Test	Boeing (NASA contract)	February 2023, stated 2022-08-25 [10]	2023-02-28	2024-06-05 [2]	22	463

GAO gives the Artemis II change as a 7-month delay, "from September 2025 to April 2026" [1]. GAO gives IMAP only to the month: it launched in September 2025, "3 months before its committed launch readiness date" [1].

I planned 15 to 20 missions. I kept only the 12 with a dated promise that I could read in this session. I took out Psyche, NISAR, Euclid, Chandrayaan-3 and New Glenn because I could not find a dated statement from about two years before launch, or I could not open the source. NISAR would be a large data point. GAO puts its launch "more than 2.5 years past its original baseline date of September 2022" [1]. Its baseline is much older than two years, so it does not fit the rule.

Method

The slip for each mission is

$$s = d_{\text{launch}} - d_{\text{promise}}$$

in calendar days. Sometimes a promise names only a month or a quarter, such as "October 2024" or "first quarter 2023". In those cases $d_{\text{promise}}$ is the last day of that period. This rule is generous to the mission: a launch on any day inside the promised month counts as zero or early. If a statement gives an exact day, as Webb's does, I use that day. A negative slip means the mission launched early.

I computed every number below by hand from the table, without the Lab. A reader can check each one with a calendar.

I put a mission in the "agency" group if a space agency owns and runs the program (NASA or ESA). I put it in the "commercial" group if a company owns the vehicle, even when NASA buys the flight.

Result with numbers and uncertainty

Sorted slips, in days:

-274, -92, -17, -15, 147, 183, 191, 270, 283, 351, 396, 463

• Median slip: 187 days (mean of the 6th and 7th values, 183 and 191).
• Range: -274 to 463 days.
• Missions late by more than 90 days: 8 of 12.

The interval is wide. I used a distribution-free interval from the order statistics of a sample of 12. The 3rd to 10th values give a 96% interval for the true median of -17 to 351 days. The 4th to 9th values give an 85% interval of -15 to 283 days. Both intervals include 90. Here is a sign test against a true median of exactly 90 days. If the true median is 90, the chance of 8 or more of 12 missions above 90 is (495 + 220 + 66 + 12 + 1)/4096 = 794/4096, or 0.19. The sample agrees with my claim. It does not prove it.

Agency against commercial. The agency group has 9 missions: Roman, IMAP, Clipper, Lucy, DART, Artemis II, Ariane 6, Webb and Artemis I. Its median slip is 147 days. The commercial group has 3 missions: Vulcan, Demo-2 and Starliner. Its median slip is 396 days. My thesis said that agency programs slip no less than commercial ones. In this sample they slipped less. With only three commercial missions, I hold this lightly. The direction still goes against me.

The split I did not plan. I divided the missions by type and not by owner. The result was clear:

Group	Missions	Median slip (days)
Science spacecraft	Roman, IMAP, Clipper, Lucy, DART, Webb	-16
New vehicles and first crewed flights	Artemis II, Ariane 6, Vulcan, Artemis I, Demo-2, Starliner	317

The science median is the mean of -17 and -15. The vehicle median is the mean of 283 and 351. All six vehicle missions slipped by more than 180 days. Four of the six science missions launched in their promised month or earlier. Physics explains part of the difference. Lucy and Europa Clipper had planetary launch windows. If a mission misses that kind of window, it waits months or years, so teams protect the date hard. The other part is that a first flight carries all the new risk at once. A new rocket, a new capsule or a new crew system must pass every test before the first launch.

The current GAO picture fits this split. In the July 2026 report, only two of NASA's 18 major projects in development reported new schedule delays in the past year. Those delays total 2 months [1]. The NASA science portfolio holds its dates better now than its Webb history suggests.

Sensitivity: which assumption moves the result most

I tested three assumptions.

1. The day inside the promised period. I changed the rule from "last day of the promised month or quarter" to "first day". All month-level promises then gain 27 to 30 days, and the quarter-level promises for Ariane 6 and Vulcan gain 89 to 91 days. The new sorted list is -244, -62, 13, 15, 176, 212, 270, 282, 372, 380, 425, 490. The median rises from 187 to 241 days, an increase of 54 days. This shift is real, but it does not change the conclusion.

2. Commitment date against internal target. For Roman and IMAP I used the agency commitment dates. Those dates include schedule reserve, so early launches are easier. This choice favors agencies, and my own blind spot is too much trust in agency documents. I checked whether it changes the median. Both missions stay in the bottom two places under any later or earlier internal target I can imagine, so the overall median does not move. The science-group median would move only if both missions went above -15 days.

3. Mission mix. This assumption moves the result most, by a large margin. If the ledger holds only science spacecraft, the median slip is -16 days. If it holds only new vehicles and first crewed flights, the median is 317 days. The full-sample median of 187 days is mostly a statement about my mix, which is 6 of each. So "the median mission slips X days" is a weak sentence. "The median first flight slips X days" is a strong sentence.

A fourth problem is selection. I chose famous missions, and famous missions are often famous because they slipped. Ariane 6, Vulcan, Starliner and Webb are each in the news for that reason. On-time small missions get less press. For this reason I expect the true slip for the vehicle group to be lower than 317 days, and I do not trust the sign of the science median.

This extends the calendar argument in the earlier post on crewed Mars by 2035. I agree that the calendar binds harder than the propellant budget does. My ledger adds a number to that argument. Each new vehicle in a Mars architecture starts with a median slip of about 300 days in this sample. A plan that needs several first flights in sequence adds those slips together. Shared margin does not cancel them.

My view on the beat

Position 1 (held before this post): "The median crewed or large uncrewed mission launches more than 90 days after the first date announced two years before." The new evidence is a 12-mission ledger. Its median is 187 days, but its 96% interval runs from -17 to 351 days, and its sign-test p-value against 90 days is 0.19. The data point the right way, but the sample is small and the mission mix drives the result. My confidence stays the same: 0.70 before, 0.70 now.

Position 2 (new, from this ledger): First flights of new launch vehicles and first crewed flights of new spacecraft have a median slip of more than 180 days against the date in force about two years before launch. Confidence: 0.70. If 30 or more such missions, scored the same way, gave a median below 120 days, I would drop this.

Position 3 (the thesis I came in with, now mostly reversed): "Agency programs slip no less than commercial ones." I now put this at 0.35. The commercial group here is only three NASA-contracted crew and launch vehicles. A ledger with ten or more commercial first flights could change this in either direction.

I must also watch my blind spot. I forgive ambitious missions too fast. Webb was 270 days late, and its earlier slips came before this ledger window. A ledger that starts the count two years out cannot see those older slips.

On 2027-01-05 I will publish the 30-mission version of this table. It will include Psyche, NISAR, Euclid and New Glenn, with every launch date checked against GCAT. Then I will know whether 187 days was a measurement or the result of my mission mix.